Deborah Hilton Statistics Online

http://sites.google.com/site/deborahhilton/

go to the favourite links page of my website and click on the electronic statistical book link and there is a useful summary there on how to deal with outliers and it talks about pairwise versus casewise deletion which is interesting as casewise can delete all your data if you have too many variables so this is good to read. let me know if you require further information - my email is on my website.

Hi Dr. Hilton, thank you very much for your reply and introducing the links on your webpage.

Best,

Talaye

Hello Talayeh, I am just Mrs, not Dr unfortunately - no more study for me. Thanks Deborah

Talayeh,

typicaly one wants to remove outliers from the data, if the goal of your model was to make inferences to a larger population. These inferences may be very unstable (at best) with the presence of outliers.

Of course, the best strategy ultimately depends on the research setting you're in. For instance, Lieberman (2005) showed how in country-comparative studies, outliers can be used to - for instance - look for omitted variables. So, he treats outliers as a starting point for future investigation, rather than as a nuisance:

Lieberman, E. S. (2005). Nested Analysis as a Mixed-Method Strategy for Comparative Research. American Political Science Review, 99, 435–452.

In addition (disclaimer: self-reference) you could look at our publication on influential data (which is related to, but not the same as, outliers). At the end of this publication we provide some suggestions on dealing with influential data that may be applicable to outliers as well.

Nieuwenhuis, R., Te Grotenhuis, M., & Pelzer, B. (2012). Influence.ME: tools for detecting influential data in mixed effects models. R Journal, 4(2), 1–10.

https://www.researchgate.net/publication/232701348_Influence.ME_tools_for_detecting_influential_data_in_mixed_effects_models

Hope this helps.

Article influence.ME: Tools for Detecting Influential Data in Mixed ...

Hi,

The very first question should be: is there an explanation as to why outliers have those values? In some cases there is something relating to a reporting anomaly (e.g. a customer paid several bills at once), and those abnormal values make sense and should be included in your analysis.

It is merely a question of whether you throw the baby out with the bathwater or not, and only a good knowedge of the data and the things you want to analyze can help. Another remark is that in the old days (30-40 years ago), in biological studies, the practice was just to delee the top 5% and bottom 5% of the observations, which is something that is not done anymore for good reasons.

Thanks for your nice answers.